@krtclcdy/tersio 2.25.4 → 2.25.6

package/dist/extensions/shared/usage-store.js

@@ -1,9 +1,9 @@
-// extensions/shared/usage-store.ts — tersio-owned usage.db. A cache of the live parse, never a fork; missing sqlite3 → sync false / read null.
-import { execFileSync } from 'node:child_process';
+// usage-store.ts — tersio-owned usage.db. A cache of the live parse, not a fork; no SQLite → sync false / read null.
 import fs from 'node:fs';
 import path from 'node:path';
-import { RECENT_LIMIT, SQLITE_READ_BUFFER, canonicalModelId, classifySessionLine, classifyOpencodeMessage, costOf, durOf, hostOfSessionFile, ingestSessionRow, messageRowId, OPENCODE_DB_OVERLAP_MS, opencodeDbPath, opencodeSessionsDir, newSessionAccum, readOpencodeDbRows, sessionsDirs, walkJsonl, } from './usage-ledger.js';
-import { tersioDataPath } from '../lib/utils.js';
+import { sqliteExec, sqliteRead } from './sqlite.js';
+import { RECENT_LIMIT, canonicalModelId, classifySessionLine, classifyOpencodeMessage, costOf, durOf, agentOfSessionText, hostOfSessionFile, ingestSessionRow, messageRowId, OPENCODE_DB_OVERLAP_MS, opencodeDbPath, opencodeSessionsDir, newSessionAccum, opencodeToolNames, readOpencodeDbRows, sessionsDirs, walkJsonl, } from './usage-ledger.js';
+import { tersioDataPath } from './utils.js';
 import { MODEL_ALIASES } from './pricing.js';
 import { tersioSettingsFile } from './plugin-settings.js';
 export function usageDbPath() {
@@ -24,25 +24,21 @@ function nullNum(v) {
 function nullStr(v) {
     return v === undefined ? 'NULL' : esc(v);
 }
+// A failed INSERT stops before COMMIT, so it leaves no DELETEs behind.
 function run(db, sql) {
-    // -bail stops the script before COMMIT, so a failed INSERT never leaves DELETEs behind.
-    execFileSync('sqlite3', ['-bail', db, sql], { stdio: ['ignore', 'ignore', 'ignore'], timeout: 30000 });
+    sqliteExec(db, sql);
 }
 function query(db, sql) {
-    const out = execFileSync('sqlite3', ['-separator', '\t', db, sql], {
-        encoding: 'utf8',
-        timeout: 30000,
-        stdio: ['ignore', 'pipe', 'ignore'],
-        maxBuffer: SQLITE_READ_BUFFER,
-    });
-    return out
+    return sqliteRead(db, sql)
         .split('\n')
         .filter((line) => line.trim())
         .map((line) => line.split('\t'));
 }
 // Bump on a parse change: unchanged transcripts are never re-read.
-const PARSER_VERSION = '11';
-// A re-parse that keeps less than half the rows means transcripts vanished mid-migration; the backup is restored instead of publishing the loss.
+const PARSER_VERSION = '13';
+/** OpenCode rows gained tool names; re-read by id, since a PARSER_VERSION bump drops rotated history. */
+const OPENCODE_TOOLS_VERSION = '1';
+// A re-parse keeping under half the rows means transcripts vanished mid-migration; the backup is restored instead.
 const REPARSE_GUARD_RATIO = 0.5;
 let guardReport = null;
 /** Why the last sync refused to publish a re-parse, if it did. */
@@ -52,9 +48,22 @@ export function reparseGuardReport() {
 const SCHEMA_SQL = `CREATE TABLE IF NOT EXISTS meta (k TEXT PRIMARY KEY, v TEXT NOT NULL);` +
     `CREATE TABLE IF NOT EXISTS files (id INTEGER PRIMARY KEY AUTOINCREMENT, path TEXT UNIQUE NOT NULL, mtime REAL NOT NULL, size INTEGER NOT NULL);` +
     `CREATE TABLE IF NOT EXISTS messages (file_id INTEGER NOT NULL REFERENCES files(id), t REAL, model TEXT NOT NULL,` +
-    ` i INTEGER NOT NULL, o INTEGER NOT NULL, d REAL, cr INTEGER NOT NULL, cw INTEGER NOT NULL,` +
-    ` usd REAL, st TEXT NOT NULL, code REAL, note TEXT, tools TEXT NOT NULL DEFAULT '[]', h TEXT, id TEXT);` +
-    `CREATE INDEX IF NOT EXISTS idx_messages_file ON messages(file_id);`;
+    ` i INTEGER NOT NULL, o INTEGER NOT NULL, d REAL, tf REAL, cr INTEGER NOT NULL, cw INTEGER NOT NULL,` +
+    ` usd REAL, st TEXT NOT NULL, code REAL, note TEXT, tools TEXT NOT NULL DEFAULT '[]', h TEXT, a TEXT, id TEXT);` +
+    // Every read filters or orders by one of these three; without them each sync and read scans the table.
+    `CREATE INDEX IF NOT EXISTS idx_messages_file ON messages(file_id);` +
+    `CREATE INDEX IF NOT EXISTS idx_messages_t ON messages(t);` +
+    `CREATE INDEX IF NOT EXISTS idx_messages_h_t ON messages(h, t);` +
+    `CREATE INDEX IF NOT EXISTS idx_messages_model ON messages(model);`;
+/** WAL + a checkpoint: the write path avoids an fsync per commit, and usage.db alone holds every row. */
+function pragmas(db) {
+    run(db, 'PRAGMA journal_mode=WAL;' +
+        'PRAGMA synchronous=NORMAL;' +
+        'PRAGMA wal_autocheckpoint=256;' +
+        'PRAGMA cache_size=-8000;' +
+        'PRAGMA temp_store=MEMORY;' +
+        'PRAGMA mmap_size=268435456;');
+}
 function storedParserVersion(db) {
     try {
         const stored = query(db, `SELECT v FROM meta WHERE k='parser_version';`);
@@ -64,6 +73,14 @@ function storedParserVersion(db) {
         return null;
     }
 }
+// A store written by a newer tersio: no forward migration, so a re-parse would swap rotated history for what is left.
+function isDowngrade(stored) {
+    if (stored === null)
+        return false;
+    const storedNum = Number(stored);
+    const currentNum = Number(PARSER_VERSION);
+    return Number.isFinite(storedNum) && Number.isFinite(currentNum) && storedNum > currentNum;
+}
 function countMessages(db) {
     try {
         const rows = query(db, `SELECT COUNT(*) FROM messages;`);
@@ -84,6 +101,16 @@ function newestOpencodeMs(db) {
         return null;
     }
 }
+/** Whether the OpenCode tools backfill ran; unfilled rows stay empty, so this widens the scan once. */
+function opencodeToolsBackfilled(db) {
+    try {
+        const rows = query(db, `SELECT v FROM meta WHERE k='opencode_tools';`);
+        return rows.length > 0 && rows[0][0] === OPENCODE_TOOLS_VERSION;
+    }
+    catch {
+        return false;
+    }
+}
 // Same pre-migration state as the newest backup: another copy churns the 3-slot rotation for nothing.
 function newestBackupMatches(stored) {
     if (stored === null)
@@ -99,7 +126,7 @@ function newestBackupMatches(stored) {
         return false;
     }
 }
-// v9 shape (path-keyed rows) migrates in place with zero re-parse, so history survives even when transcripts are gone.
+// v9 shape (path-keyed rows) migrates in place with zero re-parse, so history survives even without transcripts.
 function canMigrateInPlace(db) {
     try {
         const names = query(db, `PRAGMA table_info(messages);`).map((r) => r[1]);
@@ -116,8 +143,8 @@ function migrateInPlace(db, preCount) {
             `CREATE TABLE files_new (id INTEGER PRIMARY KEY AUTOINCREMENT, path TEXT UNIQUE NOT NULL, mtime REAL NOT NULL, size INTEGER NOT NULL);` +
             `INSERT INTO files_new (path, mtime, size) SELECT path, mtime, size FROM files;` +
             `CREATE TABLE messages_new (file_id INTEGER NOT NULL REFERENCES files_new(id), t REAL, model TEXT NOT NULL,` +
-            ` i INTEGER NOT NULL, o INTEGER NOT NULL, d REAL, cr INTEGER NOT NULL, cw INTEGER NOT NULL,` +
-            ` usd REAL, st TEXT NOT NULL, code REAL, note TEXT, tools TEXT NOT NULL DEFAULT '[]', h TEXT, id TEXT);` +
+            ` i INTEGER NOT NULL, o INTEGER NOT NULL, d REAL, tf REAL, cr INTEGER NOT NULL, cw INTEGER NOT NULL,` +
+            ` usd REAL, st TEXT NOT NULL, code REAL, note TEXT, tools TEXT NOT NULL DEFAULT '[]', h TEXT, a TEXT, id TEXT);` +
             `INSERT INTO messages_new (file_id, ${cols}) SELECT (SELECT id FROM files_new WHERE files_new.path = messages.file), ${cols} FROM messages;` +
             `DROP TABLE messages;DROP TABLE files;` +
             `ALTER TABLE files_new RENAME TO files;ALTER TABLE messages_new RENAME TO messages;` +
@@ -135,6 +162,63 @@ function migrateInPlace(db, preCount) {
         return false;
     }
 }
+// v12 adds the agent kind with a bare ALTER; existing rows read NULL and report under the main agent.
+function canAddAgent(db) {
+    try {
+        const names = query(db, `PRAGMA table_info(messages);`).map((r) => r[1]);
+        return names.includes('file_id') && !names.includes('a');
+    }
+    catch {
+        return false;
+    }
+}
+function addAgentColumn(db) {
+    try {
+        run(db, `ALTER TABLE messages ADD COLUMN a TEXT;`);
+    }
+    catch {
+        return false;
+    }
+    return true;
+}
+// v13 shape gains the first-token latency with a bare ALTER, so no row moves and no wipe runs.
+function canAddTtft(db) {
+    try {
+        const names = query(db, `PRAGMA table_info(messages);`).map((r) => r[1]);
+        return names.includes('file_id') && !names.includes('tf');
+    }
+    catch {
+        return false;
+    }
+}
+// The column is additive, and a rotated transcript cannot be re-parsed, so the ALTER precedes the backfill.
+function addTtftColumn(db) {
+    try {
+        run(db, `ALTER TABLE messages ADD COLUMN tf REAL;`);
+    }
+    catch {
+        return false;
+    }
+    return !canAddTtft(db);
+}
+/** First-token-latency marker; unchanged transcripts are never revisited, so old stores stay null. */
+const TTFT_VERSION = '1';
+function ttftBackfilled(db) {
+    try {
+        const rows = query(db, `SELECT v FROM meta WHERE k='ttft_backfill';`);
+        return rows.length > 0 && rows[0][0] === TTFT_VERSION;
+    }
+    catch {
+        return false;
+    }
+}
+/** Clears ledger mtimes so every transcript re-reads; synthetic `db#row-id` OpenCode keys stay untouched. */
+function markTranscriptsDirty(db) {
+    try {
+        run(db, `UPDATE files SET mtime = 0 WHERE path NOT LIKE '%#%';`);
+    }
+    catch { /* a store with no files table yet, or a read-only file: nothing to backfill */ }
+}
 // v10 shape gains the message id with a bare ALTER: no rows move, so no re-parse and no guard trip.
 function canAlterInPlace(db) {
     try {
@@ -154,15 +238,21 @@ function alterInPlace(db) {
     }
     return !canAlterInPlace(db);
 }
-function ensureSchema(db) {
+function ensureSchema(db, forceWipe) {
     fs.mkdirSync(path.dirname(db), { recursive: true });
     // Fold in rows stored under a pre-alias spelling; the mtime ledger will not re-read those transcripts.
     const rekeys = MODEL_ALIASES.map((a) => `UPDATE messages SET model='${a.id}' WHERE model='${a.feed}';`).join('');
     const stored = storedParserVersion(db);
+    pragmas(db);
     if (stored === PARSER_VERSION) {
         run(db, SCHEMA_SQL + rekeys);
         return;
     }
+    // A newer store keeps its rows: only an explicit force re-parse may wipe it.
+    if (!forceWipe && isDowngrade(stored)) {
+        run(db, SCHEMA_SQL + rekeys);
+        return;
+    }
     // A version bump rebuilds both tables, so no ALTER patchwork survives a migration.
     try {
         if (countMessages(db) > 0 && !newestBackupMatches(stored))
@@ -179,6 +269,16 @@ function ensureSchema(db) {
             run(db, `INSERT OR REPLACE INTO meta (k,v) VALUES ('parser_version','${PARSER_VERSION}');`);
             return;
         }
+        // The agent column is additive: every row survives and unchanged transcripts read NULL until a re-parse.
+        if (stored !== null && canAddAgent(db) && addAgentColumn(db)) {
+            run(db, `INSERT OR REPLACE INTO meta (k, v) VALUES ('parser_version','${PARSER_VERSION}');`);
+            return;
+        }
+        // The ttft column is additive too; its values come from the backfill in `syncUsageDb`.
+        if (stored !== null && canAddTtft(db) && addTtftColumn(db)) {
+            run(db, `INSERT OR REPLACE INTO meta (k, v) VALUES ('parser_version','${PARSER_VERSION}');`);
+            return;
+        }
         run(db, `DROP TABLE IF EXISTS messages;DROP TABLE IF EXISTS files;` + SCHEMA_SQL +
             `INSERT OR REPLACE INTO meta (k,v) VALUES ('parser_version','${PARSER_VERSION}');`);
     }
@@ -287,37 +387,40 @@ function readFiles(db) {
     catch { /* fresh db */ }
     return known;
 }
-// OpenCode rows carry no status, code, note, or tool list; only tokens and the msg id.
-function opencodeStoredRow(oc, host, id) {
+// OpenCode rows carry only tokens, msg id, and tools; they record as the main agent, so no NULL kind.
+function opencodeStoredRow(oc, host, id, tools) {
     return {
         t: oc.ms ?? null,
         model: canonicalModelId(oc.model),
         i: intOf(oc.usage.input),
         o: intOf(oc.usage.output),
         d: durOf(oc.durMs),
+        tf: durOf(oc.ttftMs),
         cr: intOf(oc.usage.cacheRead),
         cw: intOf(oc.usage.cacheWrite),
         usd: costOf(oc.usage),
         st: 'completed',
         code: undefined,
         note: undefined,
-        tools: [],
+        tools,
         host,
+        agent: 'main',
         id,
     };
 }
-function parseFile(text, host, file) {
+function parseFile(text, host, file, agent) {
     const rows = [];
     // Guarded: a JSONL transcript must fall through, not abort the sync.
     let oc = null;
     try {
-        oc = classifyOpencodeMessage(JSON.parse(text));
+        const message = JSON.parse(text);
+        oc = classifyOpencodeMessage(message);
+        if (oc) {
+            rows.push(opencodeStoredRow(oc, host, file ? path.basename(file).replace(/\.[^.]+$/, '') : undefined, opencodeToolNames(message)));
+            return rows;
+        }
     }
     catch { /* JSONL */ }
-    if (oc) {
-        rows.push(opencodeStoredRow(oc, host, file ? path.basename(file).replace(/\.[^.]+$/, '') : undefined));
-        return rows;
-    }
     let codexProvider = null;
     let codexModel = null;
     for (const line of text.split('\n')) {
@@ -349,6 +452,7 @@ function parseFile(text, host, file) {
                     i: intOf(usage.input),
                     o: intOf(usage.output),
                     d: undefined,
+                    tf: undefined,
                     cr: intOf(usage.cacheRead),
                     cw: intOf(usage.cacheWrite),
                     usd: costOf(usage),
@@ -357,6 +461,7 @@ function parseFile(text, host, file) {
                     note: undefined,
                     tools: [],
                     host,
+                    agent,
                     id: rid,
                 });
                 continue;
@@ -373,6 +478,7 @@ function parseFile(text, host, file) {
                 i: intOf(usage.input),
                 o: intOf(usage.output),
                 d: dur,
+                tf: durOf(parsed.ttftMs),
                 cr: intOf(usage.cacheRead),
                 cw: intOf(usage.cacheWrite),
                 usd: costOf(usage),
@@ -381,6 +487,7 @@ function parseFile(text, host, file) {
                 note,
                 tools: parsed.tools ?? [],
                 host,
+                agent,
                 id: rid,
             });
         }
@@ -389,12 +496,12 @@ function parseFile(text, host, file) {
     return rows;
 }
 function insertSql(file, r) {
-    return `INSERT INTO messages (file_id, t, model, i, o, d, cr, cw, usd, st, code, note, tools, h, id) VALUES (` +
+    return `INSERT INTO messages (file_id, t, model, i, o, d, tf, cr, cw, usd, st, code, note, tools, h, a, id) VALUES (` +
         `(SELECT id FROM files WHERE path=${esc(file)}),${r.t === null ? 'NULL' : String(r.t)},${esc(r.model)},${r.i},${r.o},` +
-        `${nullNum(r.d)},${r.cr},${r.cw},${nullNum(r.usd)},${esc(r.st)},` +
-        `${nullNum(r.code)},${nullStr(r.note)},${esc(JSON.stringify(r.tools))},${nullStr(r.host)},${nullStr(r.id)});`;
+        `${nullNum(r.d)},${nullNum(r.tf)},${r.cr},${r.cw},${nullNum(r.usd)},${esc(r.st)},` +
+        `${nullNum(r.code)},${nullStr(r.note)},${esc(JSON.stringify(r.tools))},${nullStr(r.host)},${nullStr(r.agent ?? undefined)},${nullStr(r.id)});`;
 }
-// Unchanged transcripts are skipped via mtime+size; rows for deleted ones are kept so rotation never erases history. False when sqlite3 is unavailable.
+// Unchanged transcripts are skipped via mtime+size; rows for deleted ones are kept so rotation never erases history.
 export function syncUsageDb() {
     let db;
     guardReport = null;
@@ -403,12 +510,19 @@ export function syncUsageDb() {
     try {
         db = usageDbPath();
         const stored = storedParserVersion(db);
-        migrated = stored !== null && stored !== PARSER_VERSION;
         // The operator deleted sessions on purpose and wants the mirror to match: bypass the guard.
         const forced = process.env.TERSIO_FORCE_REPARSE === '1';
+        // A downgrade keeps its rows: the wipe below cannot rebuild them from rotated transcripts.
+        const downgrade = !forced && isDowngrade(stored);
+        migrated = stored !== null && stored !== PARSER_VERSION && !downgrade;
+        if (downgrade)
+            guardReport = `store was written by a newer tersio (parser ${stored} > ${PARSER_VERSION}); kept its rows and skipped the re-parse`;
         if (migrated && !forced)
             preWipe = countMessages(db);
-        ensureSchema(db);
+        ensureSchema(db, forced);
+        // One rewrite so the additive column's NULL rows gain latency; runs while the marker is absent.
+        if (!downgrade && !ttftBackfilled(db))
+            markTranscriptsDirty(db);
         maybeScheduledBackup(db);
     }
     catch {
@@ -433,8 +547,10 @@ export function syncUsageDb() {
     const ocDbPath = opencodeDbPath();
     if (process.env.TERSIO_SESSIONS_DIR === undefined || process.env.TERSIO_OPENCODE_DB !== undefined) {
         const newest = newestOpencodeMs(db);
+        // A store predating the tool names re-reads the whole table once; the marker closes the window.
+        const since = newest === null || !opencodeToolsBackfilled(db) ? 0 : newest - OPENCODE_DB_OVERLAP_MS;
         if (newest !== null || fs.existsSync(ocDbPath)) {
-            ocDbRows.push(...readOpencodeDbRows(ocDbPath, newest === null ? 0 : newest - OPENCODE_DB_OVERLAP_MS));
+            ocDbRows.push(...readOpencodeDbRows(ocDbPath, since));
         }
     }
     const known = readFiles(db);
@@ -454,6 +570,7 @@ export function syncUsageDb() {
         if (!prev || prev.mtime !== entry.mtime || prev.size !== entry.size)
             changed.push(file);
     }
+    // `ocDbRows` is non-empty whenever the backfill window is open, so this never skips it.
     if (!changed.length && !ocDbRows.length)
         return true;
     const chunks = [];
@@ -489,7 +606,7 @@ export function syncUsageDb() {
     try {
         for (const file of changed) {
             const text = fs.readFileSync(file, 'utf8');
-            for (const r of parseFile(text, hostOfSessionFile(file), file)) {
+            for (const r of parseFile(text, hostOfSessionFile(file), file, agentOfSessionText(text))) {
                 if (!push(insertSql(file, r)))
                     return false;
             }
@@ -503,7 +620,7 @@ export function syncUsageDb() {
             if (!push(`INSERT INTO files (path, mtime, size) VALUES (${esc(key)},${row.created},0)` +
                 ` ON CONFLICT(path) DO UPDATE SET mtime=excluded.mtime;` +
                 `DELETE FROM messages WHERE file_id=(SELECT id FROM files WHERE path=${esc(key)});` +
-                insertSql(key, opencodeStoredRow(parsed, 'opencode', row.id))))
+                insertSql(key, opencodeStoredRow(parsed, 'opencode', row.id, opencodeToolNames(row.row, row.tools)))))
                 return false;
         }
     }
@@ -512,6 +629,16 @@ export function syncUsageDb() {
     }
     if (!push(`INSERT OR REPLACE INTO meta (k, v) VALUES ('last_sync',${esc(String(Date.now()))});`))
         return false;
+    // Only once the rows are written, so a sync that fails midway widens the scan again.
+    if (ocDbRows.length > 0) {
+        if (!push(`INSERT OR REPLACE INTO meta (k, v) VALUES ('opencode_tools','${OPENCODE_TOOLS_VERSION}');`))
+            return false;
+    }
+    // Set only after the rewrite commits: a crash mid-backfill leaves mtimes cleared and the marker absent.
+    if (changed.length > 0) {
+        if (!push(`INSERT OR REPLACE INTO meta (k, v) VALUES ('ttft_backfill','${TTFT_VERSION}');`))
+            return false;
+    }
     if (!flush())
         return false;
     if (preWipe > 0) {
@@ -531,6 +658,11 @@ export function syncUsageDb() {
         }
         catch { /* best-effort */ }
     }
+    // Fold the WAL back into the main file: a backup copies one file, and a checkpoint keeps it whole.
+    try {
+        run(db, `PRAGMA wal_checkpoint(TRUNCATE);`);
+    }
+    catch { /* best-effort */ }
     return true;
 }
 export function readUsageDb() {
@@ -544,21 +676,28 @@ export function readUsageDb() {
     }
     let rows;
     try {
-        rows = query(db, `SELECT t, model, i, o, d, cr, cw, usd, st, code, note, tools, h, id FROM messages;`);
+        // The wide `tools` column is read as one string and skipped in JS; a pre-agent store is read without it.
+        const names = query(db, `PRAGMA table_info(messages);`).map((r) => r[1]);
+        const agentCol = names.includes('a') ? 'a' : "''";
+        // A store predating first-token latency has no `tf` column; read NULL so the row still loads.
+        const tfCol = names.includes('tf') ? 'tf' : 'NULL';
+        rows = query(db, `SELECT t, model, i, o, d, cr, cw, usd, st, code, note, tools, h, ${agentCol}, id, ${tfCol} FROM messages;`);
     }
     catch {
         return null;
     }
     const accum = newSessionAccum();
-    for (const [t, model, i, o, d, cr, cw, usd, st, code, note, tools, h, id] of rows) {
+    for (const [t, model, i, o, d, cr, cw, usd, st, code, note, tools, h, a, id, tf] of rows) {
         const ts = t === '' ? undefined : Number(t);
         let toolNames = [];
-        try {
-            const parsed = JSON.parse(tools || '[]');
-            if (Array.isArray(parsed))
-                toolNames = parsed.filter((x) => typeof x === 'string');
+        if (tools && tools !== '[]' && tools !== '') {
+            try {
+                const parsed = JSON.parse(tools);
+                if (Array.isArray(parsed))
+                    toolNames = parsed.filter((x) => typeof x === 'string');
+            }
+            catch { /* keep empty */ }
         }
-        catch { /* keep empty */ }
         const usage = {
             input: Number(i) || 0,
             output: Number(o) || 0,
@@ -574,7 +713,7 @@ export function readUsageDb() {
             st: st === 'error' || st === 'aborted' ? st : 'completed',
             code: code === '' ? undefined : Number(code),
             note: note === '' ? undefined : note,
-        }, toolNames, h === '' ? undefined : h, id === '' || id === undefined ? undefined : id);
+        }, toolNames, h === '' ? undefined : h, id === '' || id === undefined ? undefined : id, a === '' ? undefined : a, tf === undefined || tf === '' ? undefined : Number(tf));
     }
     accum.recent.sort((a, b) => b.t - a.t);
     let syncedAt = 0;
@@ -594,6 +733,10 @@ export function readUsageDb() {
             byTool: accum.byTool,
             byModelMessages: accum.byModelMessages,
             byHost: accum.byHost,
+            byAgent: accum.byAgent,
+            byAgentMessages: accum.byAgentMessages,
+            byModelMeasured: accum.byModelMeasured,
+            byModelCostMessages: accum.byModelCostMessages,
             costMeasured: accum.costMeasured,
             recent: accum.recent.slice(0, RECENT_LIMIT),
         },

package/dist/extensions/{lib → shared}/utils.js

@@ -1,4 +1,4 @@
-// Shared helpers for the installer and the AI add-ons updater. Node built-ins only — this module must stay dependency-free.
+// Shared helpers for the installer and AI add-ons updater. Node built-ins only — must stay dependency-free.
 import { createHash } from 'node:crypto';
 import { accessSync, constants, createWriteStream, existsSync, mkdirSync, renameSync, statSync } from 'node:fs';
 import fs from 'node:fs/promises';
@@ -6,7 +6,7 @@ import https from 'node:https';
 import os from 'node:os';
 import path from 'node:path';
 // Promise.withResolvers is Node 22+; the package supports Node 18+.
-export function withResolvers() {
+function withResolvers() {
     let resolve;
     let reject;
     const promise = new Promise((res, rej) => { resolve = res; reject = rej; });
@@ -87,7 +87,7 @@ function redirectNext(res, url) {
     }
     return null;
 }
-// pi sets PI_CODING_AGENT; OMP does not. OMP does read PI_CODING_AGENT_DIR as its own relocation variable, so the override only counts under that marker.
+// pi sets PI_CODING_AGENT; OMP does not. It reads PI_CODING_AGENT_DIR to relocate, so the override only counts there.
 export function isPiProcess() {
     return process.env.PI_CODING_AGENT === 'true';
 }
@@ -95,7 +95,7 @@ export function piAgentDir() {
     const override = process.env.PI_CODING_AGENT_DIR;
     return isPiProcess() && override ? override : path.join(homeDir(), '.pi', 'agent');
 }
-// Ponytail is a Tersio dependency that npm hoists, so it can sit in Tersio's node_modules, a host's plugin dir, or beside it. Walking up covers all.
+// Ponytail is hoisted, so it sits in Tersio's node_modules, a host's plugin dir, or beside it; walking up covers all.
 export function findHoistedPackage(pkg, fromDir, ...relInside) {
     let dir = fromDir;
     for (;;) {
@@ -235,7 +235,7 @@ export async function readTextIfExists(p) {
         return null;
     }
 }
-// Chunked so a 200 MiB host binary is never held in memory. `tail` keeps the last needle.length-1 bytes of the previous chunk, because a match can begin near the end of one chunk and finish inside the next.
+// Chunked: a 200 MiB binary never fits memory. `tail` keeps needle.length-1 bytes in case a match spans chunks.
 export async function fileContains(filePath, needleText) {
     const needle = Buffer.from(needleText, 'utf8');
     if (needle.length === 0)
@@ -266,7 +266,7 @@ export async function fileContains(filePath, needleText) {
         await handle?.close().catch(() => { });
     }
 }
-// Tersio-owned data home: ~/.tersio. First use moves legacy ~/.omp/plugins/tersio-* files over so existing installs keep their history.
+// Tersio data home: ~/.tersio. First use moves legacy ~/.omp/plugins/tersio-* files over so installs keep history.
 const migratedTersioFiles = new Set();
 export function tersioHome() {
     // Matches the other TERSIO_* overrides so a test never reads the real home.

package/dist/extensions/timing/index.js

@@ -0,0 +1,31 @@
+export default function timingExtension(pi) {
+    // First token per request, keyed by the request's own start stamp, which survives every stream event.
+    const firstTokenAt = new Map();
+    pi.on('message_update', (event) => {
+        const message = event.message;
+        if (!message || message.role !== 'assistant')
+            return;
+        const anchor = typeof message.timestamp === 'number' ? message.timestamp : 0;
+        if (anchor <= 0 || firstTokenAt.has(anchor))
+            return;
+        // `start` opens the stream with no token yet; the first delta is the first token.
+        if (!String(event.assistantMessageEvent?.type ?? '').endsWith('_delta'))
+            return;
+        firstTokenAt.set(anchor, Date.now());
+    });
+    pi.on('message_end', (event) => {
+        const message = event.message;
+        if (!message || message.role !== 'assistant')
+            return;
+        const anchor = typeof message.timestamp === 'number' ? message.timestamp : 0;
+        const at = anchor > 0 ? firstTokenAt.get(anchor) : undefined;
+        if (anchor > 0)
+            firstTokenAt.delete(anchor);
+        // A host that recorded its own latency wins; this only fills the gap.
+        if (at === undefined || message.ttft !== undefined)
+            return;
+        const ttft = at - anchor;
+        if (ttft > 0)
+            message.ttft = ttft;
+    });
+}

package/dist/extensions/{ai-addons-updater → updater}/index.js

@@ -1,18 +1,26 @@
-// /ai-addons manual updater. Node built-ins only. RTK ships SHA256 checksums with no signature, so checksum-only is the best available.
+// /ai-addons manual updater. Node built-ins only. RTK ships unsigned SHA256 checksums, so checksum-only it is.
 import { execFileSync } from 'node:child_process';
 import { createHash } from 'node:crypto';
+import fsSync from 'node:fs';
 import fs from 'node:fs/promises';
 import os from 'node:os';
 import path from 'node:path';
 import { fileURLToPath } from 'node:url';
 import { setExtensionLabel } from '../shared/host.js';
-import { CAVEMAN_REMOTE_ULTRA, CAVEMAN_REMOTE_MEGACAVE, RTK_RELEASE_API, fetchJson, findFile, findHoistedPackage, httpsGet, httpsDownload, sha256Hex, parseChecksum, normalizeRtkVersion, readTextIfExists, rtkPlatformSpec, resolveRtkBinary, } from '../lib/utils.js';
+import { CAVEMAN_REMOTE_ULTRA, CAVEMAN_REMOTE_MEGACAVE, RTK_RELEASE_API, fetchJson, findFile, findHoistedPackage, httpsGet, httpsDownload, sha256Hex, parseChecksum, normalizeRtkVersion, readTextIfExists, rtkPlatformSpec, resolveRtkBinary, } from '../shared/utils.js';
 const IS_WINDOWS = process.platform === 'win32';
 const EXTENSION_DIR = path.dirname(fileURLToPath(import.meta.url));
 const PONYTAIL_REMOTE = 'https://raw.githubusercontent.com/DietrichGebert/ponytail/main/package.json';
 const RTK_BINARY = resolveRtkBinary();
-// Beside this file, not a host path: these are the rules /caveman reads.
-const cavemanRule = (name) => path.join(EXTENSION_DIR, '..', 'caveman-session', name);
+// Beside this file: a retired agent-tree copy has no `caveman` sibling, so fall back to the OMP plugin.
+const CAVEMAN_SIBLING = (name) => path.join(EXTENSION_DIR, '..', 'caveman', name);
+function cavemanRule(name) {
+    const beside = CAVEMAN_SIBLING(name);
+    if (fsSync.existsSync(beside))
+        return beside;
+    const plugin = path.join(os.homedir(), '.omp', 'plugins', 'node_modules', '@krtclcdy', 'tersio', 'extensions', 'caveman', name);
+    return fsSync.existsSync(plugin) ? plugin : beside;
+}
 const CAVEMAN_LOCAL = cavemanRule('rule.md');
 const CAVEMAN_TRACKED = [
     { file: cavemanRule('rule-ultra.md'), remote: CAVEMAN_REMOTE_ULTRA },
@@ -20,7 +28,7 @@ const CAVEMAN_TRACKED = [
 ];
 // Host-neutral: this updater runs inside pi, OMP, and OpenCode alike.
 const RELOAD_MSG = 'Reminder: restart your agent (or reload extensions) for updates to take effect.';
-// Ponytail is a hoisted dependency, so the installed copy is found by walking up rather than by naming a host directory.
+// Ponytail is hoisted, so the installed copy is found by walking up, not by naming a host directory.
 function ponytailLocal() {
     return findHoistedPackage('@dietrichgebert/ponytail', EXTENSION_DIR, 'package.json');
 }
@@ -139,7 +147,6 @@ async function checkAddons(ctx) {
     }
     return lines.join('\n');
 }
-// Ponytail is a hoisted dependency; replace its package dir from the npm tarball.
 async function updatePonytail(pi, ctx, dryRun = false) {
     void pi;
     const localPkg = ponytailLocal();

package/dist/probe-catalog.js

@@ -0,0 +1,48 @@
+// Throwaway: read the real rate cards out of the pi-catalog omp already ships.
+import { readFileSync } from "node:fs";
+const root = process.env.HOME + "/.bun/install/global/node_modules/@oh-my-pi/pi-catalog/src";
+const catalog = JSON.parse(readFileSync(root + "/models.json", "utf8"));
+const wanted = [
+    "space-bunny",
+    "muse-spark-1.3-contributor",
+    "glm-5.3-flash",
+    "qwen38-flash-next",
+    "muse-spark-1.2-contributor",
+    "qwen3.8-27b",
+    "m3",
+    "ox-alpha",
+    "deepseek-v4-flash",
+    "ling-3.0-flash-sante",
+    "qwen3.6-35b-a3b-fp8",
+    "mimo-v2.6-flash",
+    "deepseek-v4.1-flash",
+    "step-5-preview",
+    "ling-3.1-flash",
+    "longcat-2.5-preview",
+];
+let providers = 0;
+let rows = 0;
+const hits = new Map();
+for (const [provider, models] of Object.entries(catalog)) {
+    if (typeof models !== "object" || models === null)
+        continue;
+    providers += 1;
+    for (const [id, row] of Object.entries(models)) {
+        rows += 1;
+        const bare = id.toLowerCase().split("/").pop();
+        if (wanted.includes(bare)) {
+            const list = hits.get(bare) ?? [];
+            list.push({ provider, id, cost: row.cost, status: row.pricingStatus });
+            hits.set(bare, list);
+        }
+    }
+}
+console.log("providers:", providers, "model rows:", rows, "\n");
+for (const w of wanted) {
+    const list = hits.get(w) ?? [];
+    console.log(w.padEnd(26) + (list.length === 0 ? "  — NOT IN CATALOG —" : "  " + list.length + " row(s)"));
+    for (const h of list.slice(0, 4)) {
+        const c = h.cost;
+        console.log("    " + (h.provider + "/" + h.id).padEnd(52) + " in=" + String(c?.input ?? "?") + " out=" + String(c?.output ?? "?") + " status=" + JSON.stringify(h.status));
+    }
+}